知识卡片
混合权重是概率测度
有限混合p(x)=Σj wj pj(x)要求权重非负且和为1,各分量本身归一化。连续混合把求和改为对潜先验积分。权重与分量均值协方差同样属于模型定义,不能只指定分量位置就认为联合密度已经完整。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
原书出处 · 1 处
- · 查看原书依据
The mixture of Gaussians model is in fact classical model that represents a density as a weighted sum of Gaussian distributions: (33.5) where the parameters are , that is, the mean and variance of all Gaussians in the mixture. Unlike classical Gaussian mixture models, VAEs use an infinite mixture of Gaussians, that is, k → ∞. But wait, how can we parameterize an infinite mixture?
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第33章《生成建模与表示学习》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c033.xhtml)
- 结论依据:摘录中关于混合权重是概率测度的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:The mixture of Gaussians model is in fact classical model that represents a density as a weighted sum of Gaussian distributions: (33.5) where the parameters are , that is, the mean and variance of all Gaussians in the mixture. Unlike classical Gaussian mixture models, VAEs use an infinite mixture of Gaussians, that is, k → ∞. But wait, how can we parameterize an infinite mixture?